IntroductionSevere sepsis and septic shock are major reasons for inten-sive care unit (ICU) admission and leading causes of mortality

in noncoronary ICUs [1–3]. Apart from in the West, little isknown about outcomes of patients admitted to the ICU withsevere sepsis and septic shock, despite the seriousness of

R116

Critical Care October 2003 Vol 7 No 5 Arabi et al.

ResearchAssessment of six mortality prediction models in patientsadmitted with severe sepsis and septic shock to the intensivecare unit: a prospective cohort studyYaseen Arabi1, Nehad Al Shirawi1, Ziad Memish2, Srinivas Venkatesh1 and Abdullah Al-Shimemeri1

1Department of Intensive Care, King Fahad National Guard Hospital, Riyadh, Saudi Arabia2Department of Infection Prevention and Control, King Fahad National Guard Hospital, Riyadh, Saudi Arabia

Correspondence: Yaseen Arabi, yaseenarabi@yahoo.com.

APACHE = Acute Physiology and Chronic Health Evaluation; ICU = intensive care unit; MPM = Mortality Probability Model; ROC = receiver operat-ing characteristic; SAPS = Simplified Acute Physiology Score; SIRS = systemic inflammatory response syndrome; SMR = standardized mortalityratio.

Abstract

Introduction We conducted the present study to assess the validity of mortality prediction systems inpatients admitted to the intensive care unit (ICU) with severe sepsis and septic shock. We includedAcute Physiology and Health Evaluation (APACHE) II, Simplified Acute Physiology Score (SAPS) II,Mortality Probability Model (MPM) II0 and MPM II24 in our evaluation. In addition, SAPS II and MPM II24were customized for septic patients in a previous study, and the customized versions were included inthis evaluation.Materials and method This cohort, prospective, observational study was conducted in a tertiary caremedical/surgical ICU. Consecutive patients meeting the diagnostic criteria for severe sepsis and septicshock during the first 24 hours of ICU admission between March 1999 and August 2001 wereincluded. The data necessary for mortality prediction were collected prospectively as part of theongoing ICU database. Predicted and actual mortality rates, and standardized mortality ratio werecalculated. Calibration was assessed using Lemeshow–Hosmer goodness of fit C-statistic.Discrimination was assessed using receiver operating characteristic curves.Results The overall mortality prediction was adequate for all six systems because none of thestandardized mortality ratios differed significantly from 1. Calibration was inadequate for APACHE II,SAPS II, MPM II0 and MPM II24. However, the customized version of SAPS II exhibited significantlyimproved calibration (C-statistic for SAPS II 23.6 [P=0.003] and for customized SAPS II 11.5[P=0.18]). Discrimination was best for customized MPM II24 (area under the receiver operatingcharacteristic curve 0.826), followed by MPM II24 and customized SAPS II.Conclusion Although general ICU mortality system models had accurate overall mortality prediction, theyhad poor calibration. Customization of SAPS II and, to a lesser extent, MPM II24 improved calibration. Thecustomized model may be a useful tool when evaluating outcomes in patients with sepsis.

Keywords mortality, prediction, Saudi Arabia, sepsis, septic shock

Received: 25 June 2003

Revisions requested: 14 July 2003

Revisions received: 15 July 2003

Accepted: 6 August 2003

Published: 28 August 2003

Critical Care 2003, 7:R116-R122 (DOI 10.1186/cc2373)

This article is online at http://ccforum.com/content/7/5/R116

© 2003 Arabi et al., licensee BioMed Central Ltd(Print ISSN 1364-8535; Online ISSN 1466-609X). This is an OpenAccess article: verbatim copying and redistribution of this article arepermitted in all media for any purpose, provided this notice ispreserved along with the article's original URL.

Open Access

R117

sepsis as a public health problem in developing countries.According to the 1996 World Health Organization HealthReport [4], infectious and parasitic diseases caused 17 millionout of 50 million deaths globally, including 3.4 million deathsfrom acute lower respiratory infections, 3 million from tubercu-losis, 2.5 million from diarrhoeal diseases, 1.5–2.7 million frommalaria and 1.5 million from HIV/AIDS. Infectious and parasiticdiseases accounted for 43% of the 40 million deaths occur-ring in the developing countries in 1996.

With increasing international travel and the trend towardglobalization, an international perspective on the outcome fol-lowing sepsis is becoming increasingly important. TheKingdom of Saudi Arabia has some unique features in thisregard. First, the Kingdom hosts the annual Islamic pilgrimage(Hajj), when 2 million Muslims from more than 100 countriesgather in Makkah [5]. Many of these pilgrims are elderly withunderlying chronic illnesses, making them especially suscep-tible to infectious illnesses. Second, the health care system inthe Kingdom grew rapidly to state-of-the-art levels over thepast two decades, and this development has brought with itthe challenges that face modern medicine, including trans-plantation complications, cancer therapy and advancedsurgery.

Understanding sepsis outcome studies is hampered by twofactors. First is the inconsistency in the definition of sepsis.This led to a consensus statement that defined systemicinflammatory response syndrome (SIRS), sepsis, severesepsis and septic shock [6]. More recently, these definitionswere revisited; the concept of SIRS was challenged, and thedefinitions of sepsis, severe sepsis and septic shock weremaintained [7]. The second factor was the lack of an agreedseverity of illness scoring system for sepsis patients. In theabsence of such a system, it would be difficult to interpretsepsis outcome studies [8]. Mortality prediction systems havebeen introduced as tools for assessing the performance ofICUs [9–12]. Some of these systems have been customizedfor patients with specific conditions such as sepsis, and livertransplantation and long-stay ICU patients [13,14]. If thesesystems are proved to predict accurately mortality in severesepsis and septic shock, then they will have the advantage ofbeing readily available and easily incorporated into generalICU databases without additional data collection. Customizedversions of SAPS II and MPM II24 for septic patients wereintroduced by the European–North American Study of Sever-ity Systems [15]. That study included 1130 patients who metthe criteria for severe sepsis.

The objective of the present study is to assess the validity ofsix mortality prediction systems for the severe sepsis andseptic shock patient population. This was, to our knowledge,the first study of its kind to be conducted on an independentdatabase. It is also the first study to describe the outcome ofsevere sepsis and septic shock using standardized defini-tions in a non-Western country.

Materials and methodKing Fahad National Guard Hospital is a 550-bed tertiarycare teaching centre in Riyadh, Saudi Arabia. It is a transplantcentre, and it therefore receives a large number of referrals ofpatients with end-stage liver disease. The 21-bedmedical/surgical ICU has 700–800 admissions per year andis run by full-time, on-site, board-certified intensivists. The ICUdatabase was established in March 1999 to record all ICUadmissions. We included information on all consecutiveadmissions between 1 March 1999 and 31 August 2001meeting the definitions of severe sepsis and septic shock inthe first 24 hours of ICU admission. Severe sepsis is definedas the presence of sepsis associated with organ dysfunction.Septic shock is defined as sepsis-induced hypotension andperfusion abnormalities despite fluid resuscitation, necessitat-ing vasopressor support. At the time of the study, these defin-itions were based on the 1992 American College of ChestPhysicians and Society of Critical Care Medicine consensusstatement [6]. In the more recent statement, published in2003 [7], the definition of SIRS was challenged and wasreplaced by new diagnostic criteria. The definitions of severesepsis and septic shock were maintained, as mentionedabove. Therefore, the new definitions do not affect our patientpopulation.

Patients younger than 16 years, and burn and brain-deadpatients were excluded. For patients admitted to the ICUmore than once during a hospitalization episode, only datafrom the first admission were used. Approval from the hospi-tal ethics committee was not required because the informa-tion was already being collected for clinical purposes. Thefollowing data were collected: demographics, Acute Physiol-ogy and Chronic Health Evaluation (APACHE) II scores,Simplified Acute Physiology Score (SAPS) II scores, andMortality Probability Model (MPM) variables. MPM II0 datawere obtained for all admissions, whereas MPM II24,APACHE II and SAPS II data were obtained in patients whostayed 24 hours or longer in the ICU. The original methodol-ogy for data collection was followed [10–12,15]. The mainreason for ICU admission, whether the admission was fol-lowing emergency surgery, and the presence of severechronic illness were documented according to the defini-tions used in the original APACHE II article [10]. Severechronic illnesses included cirrhosis, New York Heart Associ-ation class IV heart failure, chronic respiratory failure, end-stage renal disease and immunosuppression. ICU andhospital duration of stay, and vital status at discharge bothfrom the ICU and from the hospital was documented. Hospi-tal mortality is used as the primary end-point for all mortalitypredictions.

Statistics

Predicted mortality was calculated using the logistic regres-sion formulae described in the original articles [10–12,15].The formula used for calculation of predicted mortality in thecustomized SAPS II system is as follows:

Available online http://ccforum.com/content/7/5/R116

R118

Critical Care October 2003 Vol 7 No 5 Arabi et al.

eβ0β1 (SAPS II score)

Predicted mortality =1 + eβ0β1 (SAPS II score)

where β0 is –3.5524 and β1 is 0.0694.

A similar approach was used in calculating predicted mortal-ity for the customized MPM II24:

eβ0β1 (MPM II24 logit)

Predicted mortality =1 + eβ0β1 (MPM II24 logit)

where β0 is 0.0157 and β1 is 0.7971 [15].

Standardized mortality ratio (SMR) was calculated by dividingobserved mortality by the predicted mortality. The 95% confi-dence intervals for SMRs were calculated using the observedmortality as a Poisson variable, and then dividing its 95%confidence interval by the predicted mortality [16].

System validation was tested by assessing both calibrationand discrimination values. Calibration (the ability to provide arisk estimate that corresponds to the observed mortality) wasassessed using Lemeshow–Hosmer goodness of fit C-statis-tics [17]. In order to calculate the C-statistic, the study popu-lation was divided into 10 groups of approximately equalnumbers of patients. The predicted and actual number of sur-vivors and nonsurvivors were compared statistically usingformal goodness of fit testing to determine whether the dis-crepancy between predicted and actual values was statisti-cally insignificant (P > 0.05). Discrimination was tested usingreceiver operating characteristic (ROC) curves. ROC curveswere constructed using 10% stepwise increments in pre-dicted mortality [18,19]. A comparison of the six curves wasdone by computing the areas under the ROC curves.

Continuous variables were expressed as means ± standarddeviation and were compared using standard t-test. Categori-cal values were expressed in absolute and relative frequen-cies, and were analyzed using χ2 test. P ≤ 0.05 wasconsidered statistically significant.

ResultsDemographicsDuring the period of study 250 patients met the diagnosticcriteria for severe sepsis/septic shock in the first 24 hours ofICU admission. Demographic data for these patients are sum-marized in Table 1. Of note is the high proportion of patients(54.80%) with underlying chronic illness. ICU mortality was46% (115 patients) and hospital mortality was 61%(152 patients). The differences between hospital survivorsand nonsurvivors are also shown. Nonsurvivors were older,had higher APACHE II and SAPS II scores, had shorter hos-pital duration of stay and were more likely to have chronic ill-nesses, especially liver disease and immunosuppression. The

most common source of infection was the respiratory system(41%), followed by abdominal (32%) and then urinary (17%)sites. A total of 93 patients (37%) had positive blood cul-tures, with Gram-negative bacilli being the most common(53 patients [57%]), followed by Gram-positive cocci(36 patients [39%]) and fungi (3 patients [3%]), and Gram-negative cocci (1 patient [1%]).

Predicted mortalities

Table 2 shows actual and predicted hospital mortality foreach of the six prediction systems. There was no significantdifference between the SMR for any of the systems and 1, asevident from the confidence intervals; this indicates that allthe six systems gave overall accurate mortality estimates.

Calibration

Calibration, as tested by C-statistics, was poor for the fourstandard ICU mortality prediction systems (C-statistics:MPM II0 29.79 [P < 0.001]; MPM II24 24.82 [P = 0.002];APACHE II 34.89 [P < 0.001]; SAPS II 23.60 [P = 0.003]).Customization of SAPS II and MPM II24 was associated withimprovement in calibration, but this was statistically adequateonly for customized SAPS II (C-statistic 11.48 [P = 0.18];Table 3).

Discrimination

The ROC curves are shown in Fig. 1. The correspondingareas under the ROC curves were as follows (in descendingorder to reflect decreasing levels of discrimination): cus-tomized MPM II24 0.826; MPM II24 0.823; MPM II0 0.806;customized SAPS II 0.799; SAPS II 0.797; and APACHE II0.782.

DiscussionThe main findings of this study can be summarized as follows.First, the overall mortality prediction for all six systems wasaccurate, as reflected by the SMRs. Second, calibration wasinadequate for the general (noncustomized) systems and wasimproved by customization, particularly for the SAPS II. Third,discrimination was good for all systems, especially for thecustomized versions.

Mortality risk stratification in severe sepsis and septic shockis commonly used in clinical trials and in practice [20], whichhelps to improve accuracy when evaluating new therapiesand refining indications. By facilitating comparison of theactual with predicted mortalities, the use of such systems canalso provide valuable information about the performance ofindividual ICUs in treating septic patients.

Several approaches in risk stratification have been utilized,including the use of severity of illness scoring systems andthe use of certain inflammatory markers (e.g. interleukin-1,interleukin-6, tumour necrosis factor-α) [20]. The use ofillness severity systems has several advantages, includingtheir relative simplicity and wide availability. However, there is

R119

as yet no ideal system for this group of patients. Most of thesystems were developed for general ICU patients, and whenapplied to a particular group of patients, such as those withsepsis, their accuracy declines. Customization of thesesystems to predict sepsis outcome is an attractive option.Compared with the introduction of a system specific tosepsis, the customized versions require little if any extra datacollection by units already using general ICU systems.

There are several advantages to having an internationally validmortality prediction system for patients with severe sepsisand septic shock. First, it would be useful in comparing theoutcomes of such patients between different ICUs and coun-tries. In research it would help by grouping patients in a clini-cal trial – an approach used recently in the Recombinanthuman protein C Worldwide Evaluation in Severe Sepsis(PROWESS) trial [21,22]. The use of these systems will be

Available online http://ccforum.com/content/7/5/R116

Table 1

The study population

All Survivors Nonsurvivors P

Number 250 98 152

Age (years) 58.40 ± 18.12 55.54 ± 18.96 60.24 ± 17.37 0.05

Female sex (n [%]) 112 (45) 42 (43) 70 (46) NS

APACHE II score 26.94 ± 9.45 21.69 ± 8.85 31.06 ± 7.71 <0.001

SAPS II score 55.99 ± 21.36 43.73 ± 18.41 65.69 ± 18.38 <0.001

ICU LOS (days) 10.16 ± 16.40 11.38 ± 18.75 9.20 ± 14.68 NS

Hospital LOS (days) 31.24 ± 34.69 46.82 ± 41.63 21.20 ± 24.75 <0.001

Source of admission (n [%])

Emergency room 73 (29.20) 30 (30.61) 43 (28.29) NS

Ward 151 (60.40) 50 (51.02) 101 (66.45) 0.02

Operating room 19 (7.60) 12 (12.24) 7 (4.61) 0.03

Other hospitals 7 (2.80) 6 (6.12) 1 (0.66) 0.01

Chronic illnesses (n [%])

Cirrhosis 72 (28.80) 11 (11.22) 61 (40.13) <0.001

Cardiac 12 (4.80) 6 (6.12) 6 (3.95) NS

Respiratory 9 (3.60) 5 (5.10) 4 (2.63) NS

Renal 22 (8.80) 7 (7.14) 15 (9.87) NS

Immunosuppression 34 (13.60) 7 (7.14) 27 (17.76) 0.02

Any chronic illness 137 (54.80) 35 (35.71) 102 (67.11) <0.001

APACHE, Acute Physiology and Chronic Health Evaluation; LOS, length of stay; NS, not significant; SAPS, Simplified Acute Physiology Score.

Table 2

Actual and predicted mortalities and standardized mortality ratios

Variable Number Died (n) Actual mortality Predicted mortality SMR 95% CI

MPM II0 250 152 0.61 0.55 1.10 0.99–1.21

SAPS II 208 116 0.56 0.56 1.00 0.88–1.13

MPM II24 208 116 0.56 0.61 0.92 0.80–1.03

APACHE II 208 116 0.56 0.59 0.95 0.83–1.06

Cus MPM II24 208 116 0.56 0.59 0.94 0.82–1.05

Cus SAPS II 208 116 0.56 0.56 1.00 0.87–1.12

APACHE, Acute Physiology and Chronic Health Evaluation; CI, confidence interval; Cus, customized; MPM, Mortality Prediction Model; SAPS,Simplified Acute Physiology Score; SMR, standardized mortality ratio.

R120

Critical Care October 2003 Vol 7 No 5 Arabi et al.

Table 3

Lemeshow–Hosmer goodness of fit C-statistics for the six systems

Decile n PD AD PS AS Decile n PD AD PS AS

MPM II0 APACHE II

1 25 1.45 3 23.55 22 1 21 2.40 1 18.60 20

2 25 3.78 6 21.22 19 2 21 4.98 3 16.02 18

3 25 6.45 10 18.55 15 3 21 7.21 9 13.79 12

4 25 9.60 13 15.40 12 4 21 9.93 11 11.07 10

5 25 13.47 18 11.53 7 5 21 12.49 12 8.51 9

6 25 16.97 18 8.03 7 6 20 14.06 17 5.94 3

7 25 18.81 18 6.19 7 7 20 15.95 17 4.05 3

8 25 20.53 20 4.47 5 8 20 17.08 17 2.92 3

9 25 22.69 25 2.31 0 9 20 17.94 11 2.06 9

10 25 24.29 21 0.71 4 10 20 18.91 17 1.09 3

C-statistic 29.79 P < 0.001 C-statistic 34.89 P < 0.001

MPM II24 Customized MPM II24

1 21 1.78 4 19.22 17 1 21 2.65 4 18.35 17

2 21 3.90 7 17.10 14 2 21 4.87 7 16.13 14

3 21 7.12 2 13.88 19 3 21 7.68 2 13.32 19

4 21 10.90 7 10.10 14 4 21 10.74 7 10.26 14

5 21 13.32 11 7.68 10 5 21 12.69 11 8.31 10

6 21 15.23 13 5.77 8 6 21 14.31 13 6.69 8

7 21 17.02 19 3.98 2 7 21 15.93 19 5.07 2

8 21 18.53 15 2.47 6 8 21 17.45 15 3.55 6

9 20 18.88 18 1.12 2 9 20 18.10 18 1.90 2

10 20 19.65 20 0.35 0 10 20 19.22 20 0.78 0

C-statistic 24.82 P = 0.002 C-statistic 17.51 P = 0.03

SAPS II Customized SAPS II

1 21 0.88 3 20.12 18 1 21 2.24 3 18.76 18

2 21 2.71 1 18.29 20 2 21 4.39 1 16.61 20

3 21 5.22 9 15.78 12 3 21 6.62 9 14.38 12

4 21 8.20 12 12.80 9 4 21 8.96 12 12.04 9

5 21 11.57 10 9.43 11 5 21 11.50 10 9.50 11

6 21 14.85 17 6.15 4 6 21 14.08 17 6.92 4

7 20 16.19 14 3.81 6 7 20 15.18 14 4.82 6

8 20 17.19 14 2.81 6 8 20 16.17 14 3.83 6

9 20 18.18 16 1.82 4 9 20 17.25 16 2.75 4

10 20 19.41 19 0.59 1 10 20 18.87 19 1.13 1

C-statistic 23.60 P = 0.003 C-statistic 11.48 P = 0.18

AD, actually died; APACHE, Acute Physiology and Chronic Health Evaluation; AS, actually survived; MPM, Mortality Probability Model; PD,predicted to die; PS, predicted to survive; SAPS, Simplified Acute Physiology Score.

R121

of particular value when conducting large international multi-centre studies.

There is a good body of literature addressing ICU outcomesof septic patients in Western ICUs, including USA [3], France[23], Italy [24], and the UK [25]. However, apart from in theWest, very little is reported in this field, despite the serious-ness of sepsis as a public health problem in these countries.The present study sheds some light on ICU outcomes ofpatients admitted with severe sepsis and septic shock in aMiddle Eastern country.

Our study has the strength of being prospective, using stan-dardized definitions of severe sepsis and septic shock. Thestudy also has some limitations. First, it was conducted atonly one centre. The results therefore reflect the outcome ofseptic patients in a tertiary care centre and they may not begeneralizeable to all hospitals in the country. However, thestudy gives some insight into this issue, at least from a tertiarycare perspective.

In conclusion, the present study shows that customizedversion of SAPS II (and to lesser extent the customizedMPM II24) performed well in predicting mortality in patientswith severe sepsis and septic shock. As such, the customizedversions are better options for mortality prediction in septicpatients than are the general ICU mortality prediction systems.

Competing interestsNone declared.

References1. Rangel-Frausto MS: The epidemiology of bacterial sepsis.

Infect Dis Clin North Am 1999, 13:299-312.2. Angus DC, Linde-Zwirble WT, Lidicker J, Clermont G, Carcillo J,

Pinsky MR: Epidemiology of severe sepsis in the UnitedStates: analysis of incidence, outcome, and associated costsof care. Crit Care Med 2001, 29:1303-1310.

3. Angus DC, Wax RS: Epidemiology of sepsis: an update. CritCare Med 2001, suppl:S109-S116.

4. The World Health Orgnization: The world health report archives1995–2000. [http://www.who.int/whr/2001/archives/1997/factse.htm]

5. Memish ZA, Ahmed QA: Mecca bound: the challenges ahead. JTravel Med 2002, 9:202-210.

6. Bone RC, Balk RA, Cerra FB, Dellinger RP, Fein AM, Knaus WA,Schein RM, Sibbald WJ: American College of Chest Physi-cians/Society of Critical Care Medicine Consensus Confer-ence: definitions for sepsis and organ failure and guidelinesfor the use of innovative therapies in sepsis. Chest 1992, 101:1644-1655.

7. Levy MM, Fink MP, Marshall JC, Abraham E, Angus D, Cook D,Cohen J, Opal SM, Vincent JL, Ramsay G: 2001SCCM/ESICM/ACCP/ATS/SIS International Sepsis Defini-tions Conference. Crit Care Med 2003, 31:1250-1256.

8. Friedman G, Silva E, Vincent JL: Has the mortality of septicshock changed with time? Crit Care Med 1998, 26:2078-2086.

9. Lemeshow S, Le Gall Jr: Modeling the severity of illness of ICUpatients. JAMA 1994, 272:1049-1055.

10. Knaus WA, Draper EA, Wagner DP, Zimmerman JE: APACHE II.A severity of disease classification system. Crit Care Med1985, 13:818-829.

11. Le Gall J-R, Lemeshow S, Saulnier F: A new Simplified AcutePhysiology Score (SAPS II) based on a European/NorthAmerican multi center study. JAMA 1993, 270:2957-2962.

12. Lemeshow S, Teres D, Klar J, Avrunin JS, Gehlbach SH, RapoportJ: Mortality Probability Models (MPM II) based on an interna-tional cohort of intensive care unit patients. JAMA 1993, 270:2478-2486.

13. Suistomaa M, Niskanen M, Kari A, Hynynen M, Takala J: Cus-tomized prediction models based on APACHE II and SAPS IIscores in patients with prolonged length of stay in the ICU.Intensive Care Med 2002, 28:479-485.

Available online http://ccforum.com/content/7/5/R116

Figure 1

Receiver operating characteristic (ROC) curves for the six mortality prediction systems. APACHE, Acute Physiology and Chronic Health Evaluation;cus, customized; MPM, Mortality Probability Model; SAPS, Simplified Acute Physiology Score.

R122

14. Angus DC, Clermont G, Kramer DJ, Linde-Zwirble WT, PinskyMR: Short-term and long-term outcome prediction with theAcute Physiology and Chronic Health Evaluation II in Systemafter orthotopic liver transplantation. Crit Care Med 2000, 28:150-156.

15. LeGall JR, Lemeshow S, Leleug, Klar J, Huillard J, Rui M. Teres D,Artigas A: Customized probability models for early severesepsis in adult intensive care patients. JAMA 1995, 273:644-650.

16. Goldhill DR, Sumner A: Outcome of intensive care patients in agroup of British intensive care units. Crit Care Med 1998, 26:1337-1345.

17. Lemeshow S, Hosmer DW: A Review of goodness of fit statis-tics for use in the development of logistic regression models.Am J Epidemiol 1982, 115:92-106.

18. Metz CE: Basic Principles of ROC analysis. Semin Nucl Med1978, 8:283-298.

19. Hanley JA, McNeil BJ: The meaning and use of the area undera receiver operating characteristic (ROC) curve. Radiology1982, 143:29-36.

20. Barriere SL, Lowry SF: An overview of mortality risk predictionin sepsis. Crit Care Med 1995, 23:376-393.

21. Bernard GR, Vincent JL, Laterre PF, LaRosa SP, Dhainaut JF,Lopez-Rodriguez A, Steingrub JS, Garber GE, Helterbrand JD, ElyEW, Fisher CJ Jr: Recombinant human protein C WorldwideEvaluation in Severe Sepsis (PROWESS) study group. Effi-cacy and safety of recombinant human activated protein C forsevere sepsis. N Engl J Med 2001, 344:699-709.

22. Warren HS, Suffredini AF, Eichacker PQ, Munford RS: Risks andbenefits of activated protein C treatment for severe sepsis. NEng J Med 2002, 347:1027-1030.

23. Brun-Buisson C, Doyon F, Carlet J, Dellamonica P, Gouin F, Lep-outre A, Mercier JC, Offenstadt G, Regnier B: Incidence, riskfactors, and outcome of severe sepsis and septic shock inadults. A multicenter prospective study in intensive care units.French ICU Group for Severe Sepsis. JAMA 1995, 274:968-974.

24. Salvo I, de Cian W, Musicco M, Langer M, Piadena R, Wolfler A,Montani C, Magni E: The Italian SEPSIS study: preliminaryresults on the incidence and evolution of SIRS, sepsis, severesepsis and septic shock. Intensive Care Med 1995, suppl 2:S244-S249.

25. Edbrooke DL, Hibbert CL, Kingsley JM, Smith S, Bright NM,Quinn JM: The patient-related costs of care for sepsis patientsin a United Kingdom adult general intensive care unit. CritCare Med 1999, 27:1760-1767.

Critical Care October 2003 Vol 7 No 5 Arabi et al.